Predicting treatment resistant schizophrenia
using neuromelanin-sensitive MRI, negative symptoms and age of onset
Alaya Storm 11615532
University of Amsterdam – Bachelor Psychobiology Supervisor: Marieke van der Pluijm
Abstract
Treatment resistance (TR) in schizophrenia is a major clinical problem with around a third of the patients showing poor response to antipsychotics. Clozapine has proven effect in most TR patients, however, due to dangerous side effects initiation is delayed, which reduces effectiveness. To reduce delay in effective treatment, reliable predictors are needed. Clinical variables, age of onset and negative symptoms, show to have predictive value. Further, a new MRI technique, neuromelanin-sensitive MRI (NM-MRI), indirectly measures dopamine synthesis and has the potential to be a biomarker. This study included 41 first episode
psychosis (FEP) patients and 8 healthy controls (HC) who underwent a baseline measurement, treatment response was determined at 6 month follow-up. Contrary to predictions,
neuromelanin signal in the substantia nigra did not differ between treatment responders, TR patients and HC. Further, a trend towards significance was seen for negative symptoms, with TR patients showing increased negative symptoms compared to responders. Age of onset did not differ between TR patients and responders. A binary logistic regression, including age, NM-MRI and negative symptoms, was able to classify 72.5% of the cases correctly, with a sensitivity of 46.2% and a specificity of 85.2%. Negative symptoms significantly improved the model, suggesting it to have predictive value, however, further research is needed. To more reliably predict TR in the future, more predictors need to be considered and tested in drug naive FEP patients, making results transferrable to clinical practice.
Keywords: Treatment resistant schizophrenia, FEP, neuromelanin-sensitive MRI, negative symptoms, age of onset
Introduction
Schizophrenia is a mental disorder with a life time prevalence around 1% (McGrath et al., 2008). Symptoms of schizophrenia are clustered into three domains; positive, negative and cognitive symptoms (Howes & Murray, 2014). Positive symptoms consist of feelings and behaviour that are not present in others, such as delusion, hallucinations and disorganized speech. Negative symptoms consist of a lack of feelings and behaviour that is usually present in others, such as blunt affect, anhedonia, loss of motivation and lack of social interest. Cognitive symptoms refer to the cognitive dysfunction in patients with schizophrenia and occurs in different severity and across different domains (e.g. attention, working memory, executive function) (Simpson et al., 2010). Currently, first line treatment consist of dopamine antagonist antipsychotics, however, response to these antipsychotics is heterogeneous (Levine
et al., 2012). Around 20-35% of the patients do not respond to these antipsychotics, and are classified as treatment resistant (TR) (Conley & Kelly, 2001; Demjaha et al., 2017). A patient is considered TR when symptom severity does not reduce sufficiently despite at least two adequate trials of first line antipsychotics (Suzuki et al., 2012). Without effective treatment, TR patients show decreased quality of life, increased suicide risk and adverse events
associated with both disease and treatment (Kennedy et al., 2014).
Clozapine is the only antipsychotic treatment that has proven efficacy in patients with TR (Chakos et al., 2001). However, clozapine is not given without absolute necessity due to its dangerous side effects. Clozapine can lead to agranulocytosis with risk of complication and death, however, this risk can be prevented by monitoring the white blood cell amount (Alvir et al., 1993). Further, clozapine is associated with cardiac complications, such as myocarditis and cardiomyopathy (Kilian et al., 1999). Switching TR patients to clozapine leads to an improvement in functional and clinical outcomes and a decrease in (re-)hospitalisation rates (Wheeler et al., 2008). However, due to these side effects, treatment with clozapine is often delayed, leading to a worse treatment outcome (Howes et al., 2012; Shah et al., 2018). Therefore, it is important to identify TR in an early stage to switch TR patients to clozapine faster and avoid months or years of ineffective treatment. Different clinical variables have been identified as possible predictors for TR. Two variables that repeatedly showed predictive value are negative symptoms and age of onset (Bozzatello et al., 2019; Crespo-Facorro et al., 2007; Demjaha et al., 2017). Studies into these clinical predictors found that TR patients had increased negative symptoms and a lower age of onset. Though these clinical factors have predictive value, it is important to replicate these findings and examine more predictors to accurately determine a patient’s responsiveness to treatment (Demjaha et al., 2017).
To properly identify TR in an early stage, a reliable biomarker is needed. The most well-known hypothesis for the neurobiology of schizophrenia is the dopamine hypothesis. This hypothesis states that presynaptic striatal hyperdopminergia is a shared dysfunction among patients with schizophrenia (Howes & Kapur, 2009). It has repeatedly been found that patients with schizophrenia show increased dopamine synthesis capacity in the striatum, which is thought to underlie the positive symptoms (Abi-Dargham et al., 2000; Fusar-Poli & Meyer-Lindenberg, 2012). Dopamine antagonist antipsychotics are designed to block this subcortical hyperdopaminergia and reduce symptoms. However, TR patients do not show this increased striatal dopamine synthesis capacity, as measured by [18F]DOPA positron emission tomography (PET), and instead show levels comparable to healthy controls (HC) (Demjaha et
al., 2012; Kim et al., 2016). These findings suggest a different neurobiological mechanism underlying the symptoms in TR patients and would explain the ineffectiveness of dopamine antagonist antipsychotics (Howes & Kapur, 2014). Clozapine has a low affinity for dopamine receptors and a higher affinity for other receptors (e.g. serotonergic, adrenergic) and is
effective for most TR patients (Meltzer, 1994). Thus, PET already appears to be a potential biomarker for predicting TR. However, PET is an invasive and expensive method, and thereby not suitable to implement in clinical practice.
Recently a neuromelanin-sensitive magnetic resonance imaging (NM-MRI) technique has been developed to measure dopamine activity in a non-invasive way (Sasaki et al., 2008). NM-MRI measures neuromelanin (NM), a by-product of dopamine metabolism that
accumulates with age in the substantia nigra (SN) (Xing et al., 2018; Zecca et al., 2001). This technique is more often used in Parkinson disease (Sasaki et al., 2006), but is also applicable in schizophrenia, where increased NM signal was found in responders, compared to TR patients and HC (Watanabe et al., 2014). These results are in line with studies that found increased dopamine synthesis in patients with schizophrenia (Abi-Dargham et al., 2000; Fusar-Poli & Meyer-Lindenberg, 2012) and provide evidence that NM-MRI could be used as an indirect measure of dopamine (Cassidy et al., 2019). However, this technique has yet to be applied in TR. This study will assess NM differences in the SN, measured by NM-MRI, between TR patients and responders in first episode psychosis (FEP) patients with
schizophrenia. It is hypothesized that responders show increased NM signal compared to TR patients and HC. NM signal in TR patients is expected to be comparable to HC. In addition, it is expected that NM signal in the SN adds predictive value to a prediction model when added to age of onset and negative symptoms.
Materials and methods
Participants
41 FEP patients between the age of 18 and 35 were recruited at the Early Psychosis Clinics at the AMC and at the Arkin. 8 HC, matched with patients for age, gender, IQ and smoking status, were recruited through online advertisement and advertisements at public schools or universities (HBO and MBO). Written informed consent was obtained from all participants before entering the study.
Participants were excluded when they used antipsychotics for longer than one year, had a history of dependency on other substances than nicotine or cannabis, currently used
amphetamine, cocaine, or ADHD medication, had a neurological disorder, evidence of brain damage, were unable to provide informed consent or had any MRI contra-indications
(including pregnancy). In addition, HC were excluded when they had a psychiatric disorder or used medication (except birth control).
At baseline, FEP patients underwent an NM-MRI scan, a psychiatric interview, an IQ measure, an assessment of symptoms, medication history and substance use, and filled out several questionnaires. To determine negative symptoms in FEP patients, the Positive and Negative Syndrome Scale (PANSS) was used (Kay et al., 1987). The PANSS was filled out based on the psychiatric interview and scored on a 1 to 7 scale for positive, negative and general symptoms. IQ was estimated by conducting a shortened version of the Weschler Adult Intelligence Scale (WAIS) III consisting of four subtest; Information, Block design, Arithmetic and Digit Symbol coding (Velthorst et al., 2013). Diagnoses were confirmed and age of onset determined based on the Comprehensive Assessment of Symptoms and History (CASH) (Andreasen, 1992). Cannabis use and smoking status were assessed using the
Composite International Diagnostic Interview (CIDI) (WHO, 1989). Cannabis use was scored as ‘yes’ or ‘no’ for use in the past year. Smoking was measured in number of cigarettes a day during the past year.
At 6 months follow up, a PANSS was administered to determine treatment response. Based on the remission criteria by Andreasen et al. (2005), a patient was considered to be in remission if all of the following scale items of the PANSS scored 3 or lower: delusions (P1), unusual thought content (G9), hallucinatory behaviour (P3), conceptual disorganization (P2), and mannerisms/posturing (G5). Patients were classified as TR when a prescription of at least two non-clozapine neuroleptics with adequate dose, duration and adherence, did not result in remission of these symptoms. Furthermore, a patient was considered TR when medication was switched to clozapine during follow-up period.
HC only participated in a baseline assessment which included an NM-MRI scan, an IQ measure, several questionnaires, and an interview including the Mini-International
Neuropsychiatric Interview (MINI) (Sheehan et al., 1998) to rule out the presence of other psychiatric disorders.
MRI acquisition
Data was acquired on a 3T MRI scanner (Philips, Best) with a 32 channel head coil. An MRI screening questionnaire was filled out by the participants before the scan to assure
that there were no contra-indications for the MRI. Furthermore, a drug test and, in case of women, a pregnancy test were performed using a urine test. Before entering the MRI scanning room, participants were asked to remove all metal objects. They were then placed in supine position on the MRI table and their head was fixated with cushions in the head coil to reduce movement artefacts.
A gradient recalled echo NM-MRI scan was acquired for NM signal in the SN; (TE/TR = 3.9/260 msec, FA = 40°, 8 slices, slice thickness = 2.5 mm, in‐plane resolution = 0.39 × 0.39 mm2, FOV = 162 × 199 mm, number of signal averages [NSA] = 2, with off resonance magnetization pulse, magnetization transfer frequency offset = 1200 Hz and duration = 15.6 msec). For slice placement and normalization of the NM-MRI, transversal high‐resolution structural T1‐weighted volumetric images, with full head coverage, were acquired (echo time [TE] / repetition time [TR] = 4.1/9.0 msec; 189 slices; field of view [FOV] = 284 × 284 × 170 mm; voxel size: 0.9 × 0.9 × 0.9 mm, flip angle [FA] = 8°). On these, the NM‐MRI sections were placed perpendicular to the fourth ventricle floor with coverage from the posterior commissure to halfway through the pons. For normalization, T2-weighted images were acquired (TE / TR = 331/2500 msec; 200 slices; FOV 256 × 256 × 200 mm; voxel size: 1.0 × 1.0 × 1.0 mm, FA = 90°).
NM-MRI analyses
Manual traced contrast ratio. As illustrated in Figure 1A, ITK-Snap (Yushkevich et al., 2006) was used to manually trace the SN and crus cerebri (CC), as reference area, on the MRI scans. The SN and CC were manually traced for every participant on the 3 NM-MRI slices, best projecting the SN. A contrast ratio (CR) was calculated for every participant using the signal averages from the SN and the CC; (Ssn-Scc)/Scc.
Automatic contrast ratio. NM-MRI scans were pre-processed using the latest version of Statistical Parametric Mapping (SPM 12; www.fil.ion.ucl.ac.uk). First, the NM-MRI scan and the T2-weighted scan were coregistered onto the T1. Segmentation into different tissues (grey and white matter) was performed using the T1- and T2-weighted scans. NM-MRI scans were normalized to MNI standard brain space, using DARTEL routines with a grey and white matter template based on all participants. The SN and CC were then traced on an averaged normalized NM-MRI image using ITK-Snap to create a standardized mask. The mask was then overlaid on the individual normalized NM-MRI scans to determine the average signal intensity for the SN and the CC, as shown in Figure 1B. Voxels were excluded from the mask
when the signal intensity value was over two times the SD from the mean, based on either the SN or the CC data. For every participant the automatic CR was then calculated;
(Ssn-Scc)/Scc. Figure 1
Neuromelanin-MRI scan with traced substantia nigra and crus cerebri using ITK-snap
Note. A: manual tracing on an NM-MRI scan. B: automatic traced mask on an individual normalized
NM-MRI scan. Red = substantia nigra, blue = crus cerebri Statistical analyses
First, it was tested if TR patients, responders and HC differed on age, gender,
smoking, cannabis use and IQ. Further, medication duration was tested between TR patients and responders. Differences in age, IQ and smoking were tested with an ANOVA, gender and cannabis use were tested with a chi-square and medication duration with a t-test. Assumptions were tested and when violated, non-parametric tests were used instead.
To test if there were any differences in NM-MRI contrast ratio between TR patients, responders and HC, an ANCOVA was performed with age as a covariate. Differences in age of onset and negative symptom scores on the PANSS were tested between TR patients and responders using a t-test. Furthermore, it was tested if NM-MRI, age of onset and negative symptoms had any predictive value for TR with a binary logistic regression. Assumptions were tested and when violated, non-parametric tests were used instead. A significance threshold was set at α = .05 and used throughout.
Results
Subject characteristics. Summary results and statistics are presented in Table 1. Group differences were tested for age, gender, cigarettes a day, cannabis use, IQ and medication
duration. No significant differences were found for age, gender, cigarettes a day, cannabis use and IQ. Medication duration was significantly higher for TR patients compared to responders. Table 1
Demographics characteristics
TR patients Responders Controls Statistic p-value
n = 13 n = 28 n = 8
Age 22.08 (3.30) 23.29 (4.14) 22 (3.82) χ2(2) = 0.76 .684
Gender (male/female) 10/3 24/4 7/1 .864
Cigarettes a day 7.31 (8.11) 8.14 (8.75) 2.38 (5.29) χ2(2) = 2.84 .242
Cannabis used (no/yes) 7/6 14/14 2/6 .422
IQ 89.69 (10.58) 97.25 (15.21) 97.63 (13.98) F(2,46) = 1.43 .249 Medication duration 24.25 (13.50) 12.80 (9.57) U = 88.50 .008*
Note. Mean (SD) per group. TR, treatment resistant.
*p ≤ .05
Clinical variables. Negative symptoms and age of onset were compared between TR patients and responders. Summary results and statistics are shown in Table 2. Negative symptoms were not normally distributed, thus a Mann-Whitney U test was performed. Results showed a trend towards significance, with TR patients showing increased negative symptoms compared to responders. A t-test showed that age of onset did not differ significantly between TR patients and responders.
Table 2
Note. Mean, SD and statistics for all experimental variables. CR, contrast ratio. TR, treatment resistant
Contrast ratio (CR). No assumptions were violated, thus an ANCOVA was performed for manually traced CR and for automatic CR, using age as a covariate. Summary results and statistics are shown in Table 2. For manually traced CR, one outlier was removed based on datapoint more than two times the SD from the mean. For automatic CR, one outlier was
Negative symptoms, age of onset, manually traced CR and automatic CR.
TR patients Responders Controls Statistic p-value
Mean SD Mean SD Mean SD
Negative symptoms 14.15 5.29 10.75 4.08 U = 114.50 .058
Age of onset 21.15 3.83 22.61 4.18 t(39) = -1.08 .289
Manually traced CR 0.189 0.015 0.197 0.011 0.199 0.015 F(2,44) = 1.71 .182 Automatic CR 0.171 0.012 0.178 0.012 0.172 0.015 F(2,44) = 1.39 .261
removed based on datapoint more than two times the SD from the mean and, after visual inspection, a bad fitting mask. After removing the outliers, no significant differences were found for either manually traced CR, nor automatic CR (Figure 1). Statistics including the outlier did not change the results; manually traced CR, F(2,45) = 1.71, p = .192, automatic CR, F(2,45) = 0.73, p = .489.
Figure 2
Scatterplot of Contrast Ratio for Controls, TR patients and Responders.
Note. Manually traced contrast ratio on the left, automatic traced contrast ratio on the right. Dots and
lines represent respectively separate participants and the group mean. CR, contrast ratio. TR, treatment resistant.
Predictive model. Since no assumptions were violated, a binary logistic regression was
performed in a step by step model to test the predictive value of the experimental variables for treatment response. Summary results and statistics are shown in Table 3. The model,
including age, manually traced CR and negative symptoms showed to be the best predictive model, however, not significant, χ2(3) = 7.01, p = .071. The model explained 22,4%
(Nagelkerke R2) of the variance in treatment response. Further, this model showed a sensitivity of 46.2% and a specificity of 85.2%. Overall, 72.5% of the cases were correctly classified as to belong to either TR patients or responders (Table 4).
Table 3
Binary logistic regression in a step by step model including age, CR, negative symptoms and age of onset
Step Predictor B S.E. Wald p-value R2 ΔR2
p-value Model p-value 1 Constant Age -1.037 0.078 2.103 0.093 0.243 0.712 .622 .399 .026 .388 .388 2 Constant Age CR -8.767 0.053 42.987 5.603 0.097 28.526 2.448 0.304 2.271 .118 .581 .132 .106 .120 .206 3 Constant Age CR Negative symptoms -5.412 0.054 34.893 -0.147 6.039 0.102 29.525 0.079 0.803 0.278 1.397 3.487 .370 .598 .237 .062 .224 .050* .071 4 Constant Age CR Negative symptoms Age of onset -5.790 -0.209 37.840 -0.138 0.259 6.131 0.446 30.282 0.080 0.448 0.892 0.201 1.561 2.983 0.335 .345 .654 .211 .084 .563 .234 .565 .119
Note. CR = Contrast Ratio, manually traced. *p ≤ .05
Table 4
Classification Table Treatment Response
Predicted
Observed
Group Percentage correct TR patients Responders
Group TR patients 6 7 46.2
Responders 4 23 85.2
Overall percentage 72.5
Discussion
This study examined negative symptoms, age of onset and NM-MRI signal in the SN as possible predictors for treatment response in patients with schizophrenia. Results showed no differences between TR patients and responders on both age of onset and negative symptoms. Further, there were no differences between TR patients, responders and HC for either manually traced CR nor automatic CR. The best predictive model, including age, manually traced CR and negative symptoms, explained 22.4% of the variance in treatment response and correctly classified 72,5% of the cases. The model and separate predictors were not significant. However, negative symptoms added significant improvement to the model, suggesting that negative symptoms has predictive value for treatment response.
For the clinical predictors, it was hypothesized that TR patients would show increased negative symptoms and a lower age of onset. A trend towards significance was found for negative symptoms, with TR patients showing increased negative symptoms compared to responders. Contrary to prior studies from Bozzatello et al. (2019), Crespo-Facorro et al. (2007) and Demjaha et al. (2017), age of onset did not differ between TR patients and responders,. However, it should be noted that these previous studies consisted of bigger sample sizes and included patients with various diagnoses (e.g. bipolar disorder, depression), whereas this study only included patients with schizophrenia.
In addition, it was hypothesized that NM signal would be increased for responders compared to TR patients and HC. Current results, for the manual tracing and automatic method, did not confirm this hypothesis. These findings are contrary to previous studies that found increased dopamine synthesis in responders compared to TR patients and HC (Demjaha et al., 2012; Kim et al., 2016). Further, results are not in line with prior studies that found increased NM signal in schizophrenia patients compared to HC (Watanabe et al., 2014).
Contrary to expectations, the predictive model was not significant and neither were the included variables, age, CR, negative symptoms and age of onset. However, negative
symptoms did cause a significant increase in the explained variance of the model, suggesting it to be a valuable predictor in treatment response. Furthermore, the model was still able to correctly classify a large part of the cases. The specificity was higher than the sensitivity, thus the model included more false negatives and less false positives. Therefore, it could still partly guide treatment decision and, considering the risks associated with clozapine, a model
with a decreased number of false positives could be favourable over a model with high sensitivity and lower specificity.
Comparing research into the predictors for TR schizophrenia is challenging. Looking at the studies mentioned in this paper (Bozzatello et al., 2019; Crespo-Facorro et al., 2007; Demjaha et al., 2012; Demjaha et al., 2017; Kim et al., 2016; Watanabe et al., 2014), a couple things need to be noted. First, there is no consensus about the definition of TR, which leads to studies not using the same definition for classifying patients in either the responder or TR group (Howes et al., 2017; Suzuki et al., 2011). Second, across studies there are different inclusion criteria for patients, some only include schizophrenic patients, others almost all FEP patients (e.g. bipolar disorder, psychosis not otherwise specified). Finally, use and duration of antipsychotics vary within and across these studies. These differences make it more
challenging to build a specific prediction model for medication naive, FEP patients with schizophrenia.
This study is the first to measure NM-MRI differences between TR patients and responders in schizophrenia. The biggest strength of the current study is the use of FEP patients and limited prior use of antipsychotics, which is the group that will be targeted in clinical practice by a prediction model. However, some limitations need to be noted.
First, a limitation concerning the acquisition and processing of the NM-MRI data. Reliable placement of the NM-MRI is challenging. The current study used a commonly used placement protocol, where the NM-MRI volume was placed perpendicular to the pons. However, this is challenging due to the size of the pons and the variability in anatomy between participants. Wengler et al. (2020) proposed a placement protocol to increase reliability, where the NM-MRI volume was placed along the anterior commissure-posterior commissure (AC-PC) line. Further, in this study SPM12 and DARTEL were used for
registration, which, according to Wengler et al. (2020), is not the optimal software and could lead to the mask not perfectly fitting over all participants. Further, Cassidy et al. (2019) validated NM-MRI using voxelwise analyses to measure NM signals in the SN, however, this study used CR based on averaged NM signals.
Another limitation is, although the maximum time on antipsychotics is limited in the current study, patients are on various antipsychotics for different durations of time.
Responders are mostly on the right medication and in remission, while TR patients are not, leading to an increased amount of symptoms in the latter. TR patients were also significantly
longer on antipsychotics than responders.Antipsychotics work mostly on the positive symptoms, leading to only negative symptoms as a possible predictor for TR in this study. However, a-typical antipsychotic medication, such as olanzapine, also showed to reduce negative symptoms (Altamura et al., 2002).
In addition, a-typical antipsychotics showed effect in partial responders and TR patients (Emsley et al., 2000; Conley & Kelly, 2001; Taylor & Duncan-McConnell, 2000). Since a large part of the patients in this study are on a-typical antipsychotics, this could lead to wrongfully classifying patients as responders. Further, the general consensus is that
treatment response is considered heterogeneous, therefore, classifying patients as either TR or responder is contradictory. Even after separating treatment response into TR and responders, there is lot of heterogeneity in treatment response and response trajectory within the groups (Kinon, 2019; Schennach et al., 2012). Categorizing patients as either TR or responder could lead to unreliable predictions when determining treatment response at an individual level.
This study is part of a larger study that is still including participants. It also studies other possible predictors and compares NM-MRI with PET for a subgroup of patients. Future studies should focus on gathering more data on different predictors, including biological, to make an accurate and more complete model to predict TR that has clinical relevance. It is also important to focus on studying medication naive patients since this excludes the effect of antipsychotics and gives more reliable results for implementation in clinical practice. The future of predicting treatment response with neuroimaging data is in machine learning. This powerful computational approach is useful in complex multivariate datasets. Studies using machine learning have already successfully differentiated between HC and schizophrenia patients and, to some degree, predicted treatment outcome (Arbabshirani et al., 2017; Janssen et al., 2018). Including NM-MRI data and combining it with clinical predictors could lead to a reliable prediction model in the future.
To conclude, this study did not find differences between TR patients and responders for CR or age of onset. A trend toward significance was found for increased negative symptoms in TR patients compared to responders. Further, adding negative symptoms significantly improved the prediction model, making negative symptoms a possible predictor for TR. The predictive model was not significant, however, it showed promising results for classifying cases to treatment response. Making a prediction model for TR patients with schizophrenia is challenging due to the heterogeneity in treatment response. Therefore, future
studies should focus on using a combination of biological and clinical predictors in drug naive FEP patients, to make a prediction model suitable for clinical practise.
References
Abi-Dargham, A., Rodenhiser, J., Printz, D., Zea-Ponce, Y., Gil, R., Kegeles, L., Weiss, R., Cooper, T., Mann, J., Van Heertum, R., Gorman, J. and Laruelle, M., 2000. Increased baseline occupancy of D2 receptors by dopamine in schizophrenia. Proceedings of the National Academy of Sciences, 97(14), pp.8104-8109.
Altamura, A., Velonà, I., Curreli, R., Mundo, E. and Bravi, D., 2002. Is olanzapine better than haloperidol in resistant schizophrenia? A double-blind study in partial
responders. International Journal of Psychiatry in Clinical Practice, 6(2), pp.107-111. Alvir, J., Lieberman, J., Safferman, A., Schwimmer, J. and Schaaf, J., 1993.
Clozapine-Induced Agranulocytosis -- Incidence and Risk Factors in the United States. New England Journal of Medicine, 329(3), pp.162-167.
Andreasen, N., 1992. The Comprehensive Assessment of Symptoms and History (CASH). Archives of General Psychiatry, 49(8), p.615.
Andreasen, N., Carpenter, W., Kane, J., Lasser, R., Marder, S. and Weinberger, D., 2005. Remission in Schizophrenia: Proposed Criteria and Rationale for Consensus. American Journal of Psychiatry, 162(3), pp.441-449.
Arbabshirani, M., Plis, S., Sui, J. and Calhoun, V., 2017. Single subject prediction of brain disorders in neuroimaging: Promises and pitfalls. NeuroImage, 145, pp.137-165.
Bozzatello, P., Bellino, S. and Rocca, P., 2019. Predictive Factors of Treatment Resistance in First Episode of Psychosis: A Systematic Review. Frontiers in Psychiatry, 10.
Cassidy, C., Zucca, F., Girgis, R., Baker, S., Weinstein, J., Sharp, M., Bellei, C., Valmadre, A., Vanegas, N., Kegeles, L., Brucato, G., Kang, U., Sulzer, D., Zecca, L., Abi-Dargham, A. and Horga, G., 2019. Neuromelanin-sensitive MRI as a noninvasive proxy measure of dopamine function in the human brain. Proceedings of the National Academy of
Sciences, 116(11), pp.5108-5117.
Chakos, M., Lieberman, J., Hoffman, E., Bradford, D. and Sheitman, B., 2001. Effectiveness of Second-Generation Antipsychotics in Patients With Treatment-Resistant
Schizophrenia: A Review and Meta-Analysis of Randomized Trials. American Journal of Psychiatry, 158(4), pp.518-526.
Conley, R. and Kelly, D., 2001. Management of treatment resistance in schizophrenia. Biological Psychiatry, 50(11), pp.898-911.
Crespo-Facorro, B., Pelayo-Terán, J., Pérez-Iglesias, R., Ramírez-Bonilla, M., Martínez-García, O., Pardo-Martínez-García, G. and Vázquez-Barquero, J., 2007. Predictors of acute treatment response in patients with a first episode of non-affective psychosis:
Sociodemographics, premorbid and clinical variables. Journal of Psychiatric Research, 41(8), pp.659-666.
Demjaha, A., Lappin, J., Stahl, D., Patel, M., MacCabe, J., Howes, O., Heslin, M.,
Reininghaus, U., Donoghue, K., Lomas, B., Charalambides, M., Onyejiaka, A., Fearon, P., Jones, P., Doody, G., Morgan, C., Dazzan, P. and Murray, R., 2017. Antipsychotic treatment resistance in first-episode psychosis: prevalence, subtypes and
predictors. Psychological Medicine, 47(11), pp.1981-1989.
Demjaha, A., Murray, R., McGuire, P., Kapur, S. and Howes, O., 2012. Dopamine Synthesis Capacity in Patients With Treatment-Resistant Schizophrenia. American Journal of Psychiatry, 169(11), pp.1203-1210.
Emsley, R., Raniwalla, J., Bailey, P. and Jones, A., 2000. A comparison of the effects of quetiapine (‘Seroquel’) and haloperidol in schizophrenic patients with a history of and a demonstrated, partial response to conventional antipsychotic treatment. International Clinical Psychopharmacology, 15(3), pp.121-131.
Fusar-Poli, P. and Meyer-Lindenberg, A., 2012. Striatal Presynaptic Dopamine in
Schizophrenia, Part II: Meta-Analysis of [18F/11C]-DOPA PET Studies. Schizophrenia Bulletin, 39(1), pp.33-42.
Howes, O. and Kapur, S., 2009. The Dopamine Hypothesis of Schizophrenia: Version III--The Final Common Pathway. Schizophrenia Bulletin, 35(3), pp.549-562.
Howes, O. and Kapur, S., 2014. A neurobiological hypothesis for the classification of schizophrenia: type a (hyperdopaminergic) and type B (normodopaminergic). British Journal of Psychiatry, 205(1), pp.1-3.
Howes, O. and Murray, R., 2014. Schizophrenia: an integrated sociodevelopmental-cognitive model. The Lancet, 383(9929), pp.1677-1687.
Howes, O., McCutcheon, R., Agid, O., de Bartolomeis, A., van Beveren, N., Birnbaum, M., Bloomfield, M., Bressan, R., Buchanan, R., Carpenter, W., Castle, D., Citrome, L., Daskalakis, Z., Davidson, M., Drake, R., Dursun, S., Ebdrup, B., Elkis, H., Falkai, P., Fleischacker, W., Gadelha, A., Gaughran, F., Glenthøj, B., Graff-Guerrero, A., Hallak, J., Honer, W., Kennedy, J., Kinon, B., Lawrie, S., Lee, J., Leweke, F., MacCabe, J.,
McNabb, C., Meltzer, H., Möller, H., Nakajima, S., Pantelis, C., Reis Marques, T., Remington, G., Rossell, S., Russell, B., Siu, C., Suzuki, T., Sommer, I., Taylor, D., Thomas, N., Üçok, A., Umbricht, D., Walters, J., Kane, J. and Correll, C., 2017. Treatment-Resistant Schizophrenia: Treatment Response and Resistance in Psychosis (TRRIP) Working Group Consensus Guidelines on Diagnosis and
Terminology. American Journal of Psychiatry, 174(3), pp.216-229.
Howes, O., Vergunst, F., Gee, S., McGuire, P., Kapur, S. and Taylor, D., 2012. Adherence to treatment guidelines in clinical practice: study of antipsychotic treatment prior to
clozapine initiation. British Journal of Psychiatry, 201(6), pp.481-485.
Janssen, R., Mourão-Miranda, J. and Schnack, H., 2018. Making Individual Prognoses in Psychiatry Using Neuroimaging and Machine Learning. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 3(9), pp.798-808.
Kay, S., Fiszbein, A. and Opler, L., 1987. The Positive and Negative Syndrome Scale (PANSS) for Schizophrenia. Schizophrenia Bulletin, 13(2), pp.261-276.
Kennedy, J., Altar, C., Taylor, D., Degtiar, I. and Hornberger, J., 2014. The social and economic burden of treatment-resistant schizophrenia. International Clinical Psychopharmacology, 29(2), pp.63-76.
Kilian, J., Kerr, K., Lawrence, C. and Celermajer, D., 1999. Myocarditis and cardiomyopathy associated with clozapine. The Lancet, 354(9193), pp.1841-1845.
Kim, E., Howes, O., Veronese, M., Beck, K., Seo, S., Park, J., Lee, J., Lee, Y. and Kwon, J., 2016. Presynaptic Dopamine Capacity in Patients with Treatment-Resistant
Schizophrenia Taking Clozapine: An [18F]DOPA PET Study. Neuropsychopharmacology, 42(4), pp.941-950.
Kinon, B., 2019. The Group of Treatment Resistant Schizophrenias. Heterogeneity in Treatment Resistant Schizophrenia (TRS). Frontiers in Psychiatry, 9.
Levine, S., Rabinowitz, J., Faries, D., Lawson, A. and Ascher-Svanum, H., 2012. Treatment response trajectories and antipsychotic medications: Examination of up to 18months of treatment in the CATIE chronic schizophrenia trial. Schizophrenia Research, 137(1-3), pp.141-146.
McGrath, J., Saha, S., Chant, D. and Welham, J., 2008. Schizophrenia: A Concise Overview of Incidence, Prevalence, and Mortality. Epidemiologic Reviews, 30(1), pp.67-76. Meltzer, H., 1994. An overview of the mechanism of action of clozapine. The Journal of
Clinical Psychiatry, 55(Suppl B), pp.47–52.
Sasaki, M., Shibata, E., Kudo, K. and Tohyama, K., 2008. Neuromelanin-Sensitive MRI. Clinical Neuroradiology, 18(3), pp.147-153.
Sasaki, M., Shibata, E., Tohyama, K., Takahashi, J., Otsuka, K., Tsuchiya, K., Takahashi, S., Ehara, S., Terayama, Y. and Sakai, A., 2006. Neuromelanin magnetic resonance imaging of locus ceruleus and substantia nigra in Parkinson's disease. NeuroReport, 17(11), pp.1215-1218.
Schennach, R., Meyer, S., Seemüller, F., Jäger, M., Schmauss, M., Laux, G., Pfeiffer, H., Naber, D., Schmidt, L., Gaebel, W., Klosterkötter, J., Heuser, I., Maier, W., Lemke, M., Rüther, E., Klingberg, S., Gastpar, M., Musil, R., Möller, H. and Riedel, M., 2012. Response trajectories in “real-world” naturalistically treated schizophrenia
patients. Schizophrenia Research, 139(1-3), pp.218-224.
Shah, P., Iwata, Y., Plitman, E., Brown, E., Caravaggio, F., Kim, J., Nakajima, S., Hahn, M., Remington, G., Gerretsen, P. and Graff-Guerrero, A., 2018. The impact of delay in clozapine initiation on treatment outcomes in patients with treatment-resistant schizophrenia: A systematic review. Psychiatry Research, 268, pp.114-122.
Sheehan, D., Lecrubier, Y., Sheehan, K., Amorim, P., Janavs, J., Weiller, E., Hergueta, T., Baker, R., & Dunbar, G., 1998. The Mini-International Neuropsychiatric Interview (M.I.N.I): The development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10. The Journal of Clinical Psychiatry, 59(Suppl 20), pp.22–33
Simpson, E., Kellendonk, C. and Kandel, E., 2010. A Possible Role for the Striatum in the Pathogenesis of the Cognitive Symptoms of Schizophrenia. Neuron, 65(5), pp.585-596.
Suzuki, T., Remington, G., Mulsant, B., Rajji, T., Uchida, H., Graff-Guerrero, A. and Mamo, D., 2011. Treatment resistant schizophrenia and response to antipsychotics: A
review. Schizophrenia Research, 133(1-3), pp.54-62.
Suzuki, T., Remington, G., Mulsant, B., Uchida, H., Rajji, T., Graff-Guerrero, A., Mimura, M. and Mamo, D., 2012. Defining treatment-resistant schizophrenia and response to antipsychotics: A review and recommendation. Psychiatry Research, 197(1-2), pp.1-6. Taylor, D., Taylor, D. and Duncan-McConnell, D., 2000. Refractory schizophrenia and
atypical antipsychotics. Journal of Psychopharmacology, 14(4), pp.409-418. Velthorst, E., Levine, S., Henquet, C., de Haan, L., van Os, J., Myin-Germeys, I. and
Reichenberg, A., 2013. To cut a short test even shorter: Reliability and validity of a brief assessment of intellectual ability in Schizophrenia—a control-case family
study. Cognitive Neuropsychiatry, 18(6), pp.574-593.
Watanabe, Y., Tanaka, H., Tsukabe, A., Kunitomi, Y., Nishizawa, M., Hashimoto, R., Yamamori, H., Fujimoto, M., Fukunaga, M. and Tomiyama, N., 2014. Neuromelanin Magnetic Resonance Imaging Reveals Increased Dopaminergic Neuron Activity in the Substantia Nigra of Patients with Schizophrenia. PLoS ONE, 9(8), p.e104619.
Wengler, K., He, X., Abi-Dargham, A. and Horga, G., 2020. Reproducibility assessment of neuromelanin-sensitive magnetic resonance imaging protocols for region-of-interest and voxelwise analyses. NeuroImage, 208, p.116457.
Wheeler, A., Humberstone, V. and Robinson, G., 2008. Outcomes for schizophrenia patients with clozapine treatment: how good does it get?. Journal of Psychopharmacology, 23(8), pp.957-965.
World Health Organization., 1989. Composite International Diagnostic Interview. Core Version 1.0. Geneva: WHO
Xing, Y., Sapuan, A., Dineen, R. and Auer, D., 2018. Life span pigmentation changes of the substantia nigra detected by neuromelanin-sensitive MRI. Movement Disorders, 33(11), pp.1792-1799.
Yushkevich, P., Piven, J., Hazlett, H., Smith, R., Ho, S., Gee, J. and Gerig, G., 2006. User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability. NeuroImage, 31(3), pp.1116-1128.
Zecca, L., Gallorini, M., Schünemann, V., Trautwein, A., Gerlach, M., Riederer, P., Vezzoni, P. and Tampellini, D., 2001. Iron, neuromelanin and ferritin content in the substantia nigra of normal subjects at different ages: consequences for iron storage and
